Application Of Image Segmentation In Free Surface Recognition

The sloshing phenomena widely exists in the areas of ocean engineering and aerospace engineering.Model experimental investigation is an important means to verify theoretical simulation,numerical calculation and the physical characteristics of fluid.The free surface recognition is the premise to analyze hydrodynamics and how to accurately extract the free surface is a technical problem in experiment of liquid dynamics all the time.This paper attempts to use image segmentation technology to solve the problem of free surface segmentation and expects to have certain application value in the filed of the surface extraction.When the excitation frequency is closed to the natural frequency of the liqiud in the tank,the violent free surface may break and the wave tip entraps the air,which impact with gas-entrapment may cause localized and large load,which may lead to structural failure.Investigate the natural frequencies is part of the analysis of the gas-liquid coupled problem.The main contents of this paper are as follows:In this paper,image segmentation technology is used to carry out a series of research on the recognition of free surface in tank.The result of the evaluating merics for free surface segmentation:watershed algorithm based on the difference of Gaussian filter accurately recognize the smooth free surface,whose accuracy is over 99%.However,the algorithm is a certain deviation in the recognition of the free surface with blurred and partially broken.Watershed algorithm based on Sobel operation filtering can recognize free surface with local breakup,blurring and smiling bubbles,whose accuracy is over 98%.However,there are some limitations for the broken free surface.In this paper,an effective model of breaking wave based on U-net network can recognize the broken free surface.As a consequence,the segmentation technology has good stability and high accuracy in the broken free surface segmentation,whose accuracy is over 95%.Intelligent recognition technology based on U-net network model can not only make up for the shortcomings of traditional algorithms,but also ensure the accuracy of recognition.Therefore,the intelligent recognition technology based on U-net network model has wide application prospects in hydrodynamic experiments.By using Hilbert transform method based on EMD,this paper analyzes the natural frequency and other parameters of the liquid in the tank.The original signal was decomposed by EMD to remove the noise and other interfering signals,and the natural frequency and damping were obtained by Hilbert transform.Comparative analysis shows that the frist-order natural frequency of the original signal solved is smaller than the theoretical value,about0.8-0.9 f₁.Because of the noise reduction effect of EMD method,the frist-order natural frequency is smaller than the theoretical value of the Fast Fourier transformation.The damping of IMF component increases with the decrease of IMF correlation.